1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2018-2020 Intel Corporation.
4 * Copyright (C) 2020 Red Hat, Inc.
5 *
6 * Author: Tiwei Bie <tiwei.bie@intel.com>
7 * Jason Wang <jasowang@redhat.com>
8 *
9 * Thanks Michael S. Tsirkin for the valuable comments and
10 * suggestions. And thanks to Cunming Liang and Zhihong Wang for all
11 * their supports.
12 */
13
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/cdev.h>
17 #include <linux/device.h>
18 #include <linux/mm.h>
19 #include <linux/slab.h>
20 #include <linux/iommu.h>
21 #include <linux/uuid.h>
22 #include <linux/vdpa.h>
23 #include <linux/nospec.h>
24 #include <linux/vhost.h>
25
26 #include "vhost.h"
27
28 enum {
29 VHOST_VDPA_BACKEND_FEATURES =
30 (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) |
31 (1ULL << VHOST_BACKEND_F_IOTLB_BATCH) |
32 (1ULL << VHOST_BACKEND_F_IOTLB_ASID),
33 };
34
35 #define VHOST_VDPA_DEV_MAX (1U << MINORBITS)
36
37 #define VHOST_VDPA_IOTLB_BUCKETS 16
38
39 struct vhost_vdpa_as {
40 struct hlist_node hash_link;
41 struct vhost_iotlb iotlb;
42 u32 id;
43 };
44
45 struct vhost_vdpa {
46 struct vhost_dev vdev;
47 struct iommu_domain *domain;
48 struct vhost_virtqueue *vqs;
49 struct completion completion;
50 struct vdpa_device *vdpa;
51 struct hlist_head as[VHOST_VDPA_IOTLB_BUCKETS];
52 struct device dev;
53 struct cdev cdev;
54 atomic_t opened;
55 u32 nvqs;
56 int virtio_id;
57 int minor;
58 struct eventfd_ctx *config_ctx;
59 int in_batch;
60 struct vdpa_iova_range range;
61 u32 batch_asid;
62 };
63
64 static DEFINE_IDA(vhost_vdpa_ida);
65
66 static dev_t vhost_vdpa_major;
67
iotlb_to_asid(struct vhost_iotlb * iotlb)68 static inline u32 iotlb_to_asid(struct vhost_iotlb *iotlb)
69 {
70 struct vhost_vdpa_as *as = container_of(iotlb, struct
71 vhost_vdpa_as, iotlb);
72 return as->id;
73 }
74
asid_to_as(struct vhost_vdpa * v,u32 asid)75 static struct vhost_vdpa_as *asid_to_as(struct vhost_vdpa *v, u32 asid)
76 {
77 struct hlist_head *head = &v->as[asid % VHOST_VDPA_IOTLB_BUCKETS];
78 struct vhost_vdpa_as *as;
79
80 hlist_for_each_entry(as, head, hash_link)
81 if (as->id == asid)
82 return as;
83
84 return NULL;
85 }
86
asid_to_iotlb(struct vhost_vdpa * v,u32 asid)87 static struct vhost_iotlb *asid_to_iotlb(struct vhost_vdpa *v, u32 asid)
88 {
89 struct vhost_vdpa_as *as = asid_to_as(v, asid);
90
91 if (!as)
92 return NULL;
93
94 return &as->iotlb;
95 }
96
vhost_vdpa_alloc_as(struct vhost_vdpa * v,u32 asid)97 static struct vhost_vdpa_as *vhost_vdpa_alloc_as(struct vhost_vdpa *v, u32 asid)
98 {
99 struct hlist_head *head = &v->as[asid % VHOST_VDPA_IOTLB_BUCKETS];
100 struct vhost_vdpa_as *as;
101
102 if (asid_to_as(v, asid))
103 return NULL;
104
105 if (asid >= v->vdpa->nas)
106 return NULL;
107
108 as = kmalloc(sizeof(*as), GFP_KERNEL);
109 if (!as)
110 return NULL;
111
112 vhost_iotlb_init(&as->iotlb, 0, 0);
113 as->id = asid;
114 hlist_add_head(&as->hash_link, head);
115
116 return as;
117 }
118
vhost_vdpa_find_alloc_as(struct vhost_vdpa * v,u32 asid)119 static struct vhost_vdpa_as *vhost_vdpa_find_alloc_as(struct vhost_vdpa *v,
120 u32 asid)
121 {
122 struct vhost_vdpa_as *as = asid_to_as(v, asid);
123
124 if (as)
125 return as;
126
127 return vhost_vdpa_alloc_as(v, asid);
128 }
129
vhost_vdpa_remove_as(struct vhost_vdpa * v,u32 asid)130 static int vhost_vdpa_remove_as(struct vhost_vdpa *v, u32 asid)
131 {
132 struct vhost_vdpa_as *as = asid_to_as(v, asid);
133
134 if (!as)
135 return -EINVAL;
136
137 hlist_del(&as->hash_link);
138 vhost_iotlb_reset(&as->iotlb);
139 kfree(as);
140
141 return 0;
142 }
143
handle_vq_kick(struct vhost_work * work)144 static void handle_vq_kick(struct vhost_work *work)
145 {
146 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
147 poll.work);
148 struct vhost_vdpa *v = container_of(vq->dev, struct vhost_vdpa, vdev);
149 const struct vdpa_config_ops *ops = v->vdpa->config;
150
151 ops->kick_vq(v->vdpa, vq - v->vqs);
152 }
153
vhost_vdpa_virtqueue_cb(void * private)154 static irqreturn_t vhost_vdpa_virtqueue_cb(void *private)
155 {
156 struct vhost_virtqueue *vq = private;
157 struct eventfd_ctx *call_ctx = vq->call_ctx.ctx;
158
159 if (call_ctx)
160 eventfd_signal(call_ctx, 1);
161
162 return IRQ_HANDLED;
163 }
164
vhost_vdpa_config_cb(void * private)165 static irqreturn_t vhost_vdpa_config_cb(void *private)
166 {
167 struct vhost_vdpa *v = private;
168 struct eventfd_ctx *config_ctx = v->config_ctx;
169
170 if (config_ctx)
171 eventfd_signal(config_ctx, 1);
172
173 return IRQ_HANDLED;
174 }
175
vhost_vdpa_setup_vq_irq(struct vhost_vdpa * v,u16 qid)176 static void vhost_vdpa_setup_vq_irq(struct vhost_vdpa *v, u16 qid)
177 {
178 struct vhost_virtqueue *vq = &v->vqs[qid];
179 const struct vdpa_config_ops *ops = v->vdpa->config;
180 struct vdpa_device *vdpa = v->vdpa;
181 int ret, irq;
182
183 if (!ops->get_vq_irq)
184 return;
185
186 irq = ops->get_vq_irq(vdpa, qid);
187 if (irq < 0)
188 return;
189
190 irq_bypass_unregister_producer(&vq->call_ctx.producer);
191 if (!vq->call_ctx.ctx)
192 return;
193
194 vq->call_ctx.producer.token = vq->call_ctx.ctx;
195 vq->call_ctx.producer.irq = irq;
196 ret = irq_bypass_register_producer(&vq->call_ctx.producer);
197 if (unlikely(ret))
198 dev_info(&v->dev, "vq %u, irq bypass producer (token %p) registration fails, ret = %d\n",
199 qid, vq->call_ctx.producer.token, ret);
200 }
201
vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa * v,u16 qid)202 static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid)
203 {
204 struct vhost_virtqueue *vq = &v->vqs[qid];
205
206 irq_bypass_unregister_producer(&vq->call_ctx.producer);
207 }
208
vhost_vdpa_reset(struct vhost_vdpa * v)209 static int vhost_vdpa_reset(struct vhost_vdpa *v)
210 {
211 struct vdpa_device *vdpa = v->vdpa;
212
213 v->in_batch = 0;
214
215 return vdpa_reset(vdpa);
216 }
217
vhost_vdpa_get_device_id(struct vhost_vdpa * v,u8 __user * argp)218 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
219 {
220 struct vdpa_device *vdpa = v->vdpa;
221 const struct vdpa_config_ops *ops = vdpa->config;
222 u32 device_id;
223
224 device_id = ops->get_device_id(vdpa);
225
226 if (copy_to_user(argp, &device_id, sizeof(device_id)))
227 return -EFAULT;
228
229 return 0;
230 }
231
vhost_vdpa_get_status(struct vhost_vdpa * v,u8 __user * statusp)232 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp)
233 {
234 struct vdpa_device *vdpa = v->vdpa;
235 const struct vdpa_config_ops *ops = vdpa->config;
236 u8 status;
237
238 status = ops->get_status(vdpa);
239
240 if (copy_to_user(statusp, &status, sizeof(status)))
241 return -EFAULT;
242
243 return 0;
244 }
245
vhost_vdpa_set_status(struct vhost_vdpa * v,u8 __user * statusp)246 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
247 {
248 struct vdpa_device *vdpa = v->vdpa;
249 const struct vdpa_config_ops *ops = vdpa->config;
250 u8 status, status_old;
251 u32 nvqs = v->nvqs;
252 int ret;
253 u16 i;
254
255 if (copy_from_user(&status, statusp, sizeof(status)))
256 return -EFAULT;
257
258 status_old = ops->get_status(vdpa);
259
260 /*
261 * Userspace shouldn't remove status bits unless reset the
262 * status to 0.
263 */
264 if (status != 0 && (status_old & ~status) != 0)
265 return -EINVAL;
266
267 if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK))
268 for (i = 0; i < nvqs; i++)
269 vhost_vdpa_unsetup_vq_irq(v, i);
270
271 if (status == 0) {
272 ret = vdpa_reset(vdpa);
273 if (ret)
274 return ret;
275 } else
276 vdpa_set_status(vdpa, status);
277
278 if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK))
279 for (i = 0; i < nvqs; i++)
280 vhost_vdpa_setup_vq_irq(v, i);
281
282 return 0;
283 }
284
vhost_vdpa_config_validate(struct vhost_vdpa * v,struct vhost_vdpa_config * c)285 static int vhost_vdpa_config_validate(struct vhost_vdpa *v,
286 struct vhost_vdpa_config *c)
287 {
288 struct vdpa_device *vdpa = v->vdpa;
289 size_t size = vdpa->config->get_config_size(vdpa);
290
291 if (c->len == 0 || c->off > size)
292 return -EINVAL;
293
294 if (c->len > size - c->off)
295 return -E2BIG;
296
297 return 0;
298 }
299
vhost_vdpa_get_config(struct vhost_vdpa * v,struct vhost_vdpa_config __user * c)300 static long vhost_vdpa_get_config(struct vhost_vdpa *v,
301 struct vhost_vdpa_config __user *c)
302 {
303 struct vdpa_device *vdpa = v->vdpa;
304 struct vhost_vdpa_config config;
305 unsigned long size = offsetof(struct vhost_vdpa_config, buf);
306 u8 *buf;
307
308 if (copy_from_user(&config, c, size))
309 return -EFAULT;
310 if (vhost_vdpa_config_validate(v, &config))
311 return -EINVAL;
312 buf = kvzalloc(config.len, GFP_KERNEL);
313 if (!buf)
314 return -ENOMEM;
315
316 vdpa_get_config(vdpa, config.off, buf, config.len);
317
318 if (copy_to_user(c->buf, buf, config.len)) {
319 kvfree(buf);
320 return -EFAULT;
321 }
322
323 kvfree(buf);
324 return 0;
325 }
326
vhost_vdpa_set_config(struct vhost_vdpa * v,struct vhost_vdpa_config __user * c)327 static long vhost_vdpa_set_config(struct vhost_vdpa *v,
328 struct vhost_vdpa_config __user *c)
329 {
330 struct vdpa_device *vdpa = v->vdpa;
331 struct vhost_vdpa_config config;
332 unsigned long size = offsetof(struct vhost_vdpa_config, buf);
333 u8 *buf;
334
335 if (copy_from_user(&config, c, size))
336 return -EFAULT;
337 if (vhost_vdpa_config_validate(v, &config))
338 return -EINVAL;
339
340 buf = vmemdup_user(c->buf, config.len);
341 if (IS_ERR(buf))
342 return PTR_ERR(buf);
343
344 vdpa_set_config(vdpa, config.off, buf, config.len);
345
346 kvfree(buf);
347 return 0;
348 }
349
vhost_vdpa_can_suspend(const struct vhost_vdpa * v)350 static bool vhost_vdpa_can_suspend(const struct vhost_vdpa *v)
351 {
352 struct vdpa_device *vdpa = v->vdpa;
353 const struct vdpa_config_ops *ops = vdpa->config;
354
355 return ops->suspend;
356 }
357
vhost_vdpa_get_features(struct vhost_vdpa * v,u64 __user * featurep)358 static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep)
359 {
360 struct vdpa_device *vdpa = v->vdpa;
361 const struct vdpa_config_ops *ops = vdpa->config;
362 u64 features;
363
364 features = ops->get_device_features(vdpa);
365
366 if (copy_to_user(featurep, &features, sizeof(features)))
367 return -EFAULT;
368
369 return 0;
370 }
371
vhost_vdpa_set_features(struct vhost_vdpa * v,u64 __user * featurep)372 static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep)
373 {
374 struct vdpa_device *vdpa = v->vdpa;
375 const struct vdpa_config_ops *ops = vdpa->config;
376 u64 features;
377
378 /*
379 * It's not allowed to change the features after they have
380 * been negotiated.
381 */
382 if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK)
383 return -EBUSY;
384
385 if (copy_from_user(&features, featurep, sizeof(features)))
386 return -EFAULT;
387
388 if (vdpa_set_features(vdpa, features))
389 return -EINVAL;
390
391 return 0;
392 }
393
vhost_vdpa_get_vring_num(struct vhost_vdpa * v,u16 __user * argp)394 static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp)
395 {
396 struct vdpa_device *vdpa = v->vdpa;
397 const struct vdpa_config_ops *ops = vdpa->config;
398 u16 num;
399
400 num = ops->get_vq_num_max(vdpa);
401
402 if (copy_to_user(argp, &num, sizeof(num)))
403 return -EFAULT;
404
405 return 0;
406 }
407
vhost_vdpa_config_put(struct vhost_vdpa * v)408 static void vhost_vdpa_config_put(struct vhost_vdpa *v)
409 {
410 if (v->config_ctx) {
411 eventfd_ctx_put(v->config_ctx);
412 v->config_ctx = NULL;
413 }
414 }
415
vhost_vdpa_set_config_call(struct vhost_vdpa * v,u32 __user * argp)416 static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp)
417 {
418 struct vdpa_callback cb;
419 int fd;
420 struct eventfd_ctx *ctx;
421
422 cb.callback = vhost_vdpa_config_cb;
423 cb.private = v;
424 if (copy_from_user(&fd, argp, sizeof(fd)))
425 return -EFAULT;
426
427 ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
428 swap(ctx, v->config_ctx);
429
430 if (!IS_ERR_OR_NULL(ctx))
431 eventfd_ctx_put(ctx);
432
433 if (IS_ERR(v->config_ctx)) {
434 long ret = PTR_ERR(v->config_ctx);
435
436 v->config_ctx = NULL;
437 return ret;
438 }
439
440 v->vdpa->config->set_config_cb(v->vdpa, &cb);
441
442 return 0;
443 }
444
vhost_vdpa_get_iova_range(struct vhost_vdpa * v,u32 __user * argp)445 static long vhost_vdpa_get_iova_range(struct vhost_vdpa *v, u32 __user *argp)
446 {
447 struct vhost_vdpa_iova_range range = {
448 .first = v->range.first,
449 .last = v->range.last,
450 };
451
452 if (copy_to_user(argp, &range, sizeof(range)))
453 return -EFAULT;
454 return 0;
455 }
456
vhost_vdpa_get_config_size(struct vhost_vdpa * v,u32 __user * argp)457 static long vhost_vdpa_get_config_size(struct vhost_vdpa *v, u32 __user *argp)
458 {
459 struct vdpa_device *vdpa = v->vdpa;
460 const struct vdpa_config_ops *ops = vdpa->config;
461 u32 size;
462
463 size = ops->get_config_size(vdpa);
464
465 if (copy_to_user(argp, &size, sizeof(size)))
466 return -EFAULT;
467
468 return 0;
469 }
470
vhost_vdpa_get_vqs_count(struct vhost_vdpa * v,u32 __user * argp)471 static long vhost_vdpa_get_vqs_count(struct vhost_vdpa *v, u32 __user *argp)
472 {
473 struct vdpa_device *vdpa = v->vdpa;
474
475 if (copy_to_user(argp, &vdpa->nvqs, sizeof(vdpa->nvqs)))
476 return -EFAULT;
477
478 return 0;
479 }
480
481 /* After a successful return of ioctl the device must not process more
482 * virtqueue descriptors. The device can answer to read or writes of config
483 * fields as if it were not suspended. In particular, writing to "queue_enable"
484 * with a value of 1 will not make the device start processing buffers.
485 */
vhost_vdpa_suspend(struct vhost_vdpa * v)486 static long vhost_vdpa_suspend(struct vhost_vdpa *v)
487 {
488 struct vdpa_device *vdpa = v->vdpa;
489 const struct vdpa_config_ops *ops = vdpa->config;
490
491 if (!ops->suspend)
492 return -EOPNOTSUPP;
493
494 return ops->suspend(vdpa);
495 }
496
vhost_vdpa_vring_ioctl(struct vhost_vdpa * v,unsigned int cmd,void __user * argp)497 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
498 void __user *argp)
499 {
500 struct vdpa_device *vdpa = v->vdpa;
501 const struct vdpa_config_ops *ops = vdpa->config;
502 struct vdpa_vq_state vq_state;
503 struct vdpa_callback cb;
504 struct vhost_virtqueue *vq;
505 struct vhost_vring_state s;
506 u32 idx;
507 long r;
508
509 r = get_user(idx, (u32 __user *)argp);
510 if (r < 0)
511 return r;
512
513 if (idx >= v->nvqs)
514 return -ENOBUFS;
515
516 idx = array_index_nospec(idx, v->nvqs);
517 vq = &v->vqs[idx];
518
519 switch (cmd) {
520 case VHOST_VDPA_SET_VRING_ENABLE:
521 if (copy_from_user(&s, argp, sizeof(s)))
522 return -EFAULT;
523 ops->set_vq_ready(vdpa, idx, s.num);
524 return 0;
525 case VHOST_VDPA_GET_VRING_GROUP:
526 if (!ops->get_vq_group)
527 return -EOPNOTSUPP;
528 s.index = idx;
529 s.num = ops->get_vq_group(vdpa, idx);
530 if (s.num >= vdpa->ngroups)
531 return -EIO;
532 else if (copy_to_user(argp, &s, sizeof(s)))
533 return -EFAULT;
534 return 0;
535 case VHOST_VDPA_SET_GROUP_ASID:
536 if (copy_from_user(&s, argp, sizeof(s)))
537 return -EFAULT;
538 if (s.num >= vdpa->nas)
539 return -EINVAL;
540 if (!ops->set_group_asid)
541 return -EOPNOTSUPP;
542 return ops->set_group_asid(vdpa, idx, s.num);
543 case VHOST_GET_VRING_BASE:
544 r = ops->get_vq_state(v->vdpa, idx, &vq_state);
545 if (r)
546 return r;
547
548 vq->last_avail_idx = vq_state.split.avail_index;
549 break;
550 }
551
552 r = vhost_vring_ioctl(&v->vdev, cmd, argp);
553 if (r)
554 return r;
555
556 switch (cmd) {
557 case VHOST_SET_VRING_ADDR:
558 if (ops->set_vq_address(vdpa, idx,
559 (u64)(uintptr_t)vq->desc,
560 (u64)(uintptr_t)vq->avail,
561 (u64)(uintptr_t)vq->used))
562 r = -EINVAL;
563 break;
564
565 case VHOST_SET_VRING_BASE:
566 vq_state.split.avail_index = vq->last_avail_idx;
567 if (ops->set_vq_state(vdpa, idx, &vq_state))
568 r = -EINVAL;
569 break;
570
571 case VHOST_SET_VRING_CALL:
572 if (vq->call_ctx.ctx) {
573 cb.callback = vhost_vdpa_virtqueue_cb;
574 cb.private = vq;
575 } else {
576 cb.callback = NULL;
577 cb.private = NULL;
578 }
579 ops->set_vq_cb(vdpa, idx, &cb);
580 vhost_vdpa_setup_vq_irq(v, idx);
581 break;
582
583 case VHOST_SET_VRING_NUM:
584 ops->set_vq_num(vdpa, idx, vq->num);
585 break;
586 }
587
588 return r;
589 }
590
vhost_vdpa_unlocked_ioctl(struct file * filep,unsigned int cmd,unsigned long arg)591 static long vhost_vdpa_unlocked_ioctl(struct file *filep,
592 unsigned int cmd, unsigned long arg)
593 {
594 struct vhost_vdpa *v = filep->private_data;
595 struct vhost_dev *d = &v->vdev;
596 void __user *argp = (void __user *)arg;
597 u64 __user *featurep = argp;
598 u64 features;
599 long r = 0;
600
601 if (cmd == VHOST_SET_BACKEND_FEATURES) {
602 if (copy_from_user(&features, featurep, sizeof(features)))
603 return -EFAULT;
604 if (features & ~(VHOST_VDPA_BACKEND_FEATURES |
605 BIT_ULL(VHOST_BACKEND_F_SUSPEND)))
606 return -EOPNOTSUPP;
607 if ((features & BIT_ULL(VHOST_BACKEND_F_SUSPEND)) &&
608 !vhost_vdpa_can_suspend(v))
609 return -EOPNOTSUPP;
610 vhost_set_backend_features(&v->vdev, features);
611 return 0;
612 }
613
614 mutex_lock(&d->mutex);
615
616 switch (cmd) {
617 case VHOST_VDPA_GET_DEVICE_ID:
618 r = vhost_vdpa_get_device_id(v, argp);
619 break;
620 case VHOST_VDPA_GET_STATUS:
621 r = vhost_vdpa_get_status(v, argp);
622 break;
623 case VHOST_VDPA_SET_STATUS:
624 r = vhost_vdpa_set_status(v, argp);
625 break;
626 case VHOST_VDPA_GET_CONFIG:
627 r = vhost_vdpa_get_config(v, argp);
628 break;
629 case VHOST_VDPA_SET_CONFIG:
630 r = vhost_vdpa_set_config(v, argp);
631 break;
632 case VHOST_GET_FEATURES:
633 r = vhost_vdpa_get_features(v, argp);
634 break;
635 case VHOST_SET_FEATURES:
636 r = vhost_vdpa_set_features(v, argp);
637 break;
638 case VHOST_VDPA_GET_VRING_NUM:
639 r = vhost_vdpa_get_vring_num(v, argp);
640 break;
641 case VHOST_VDPA_GET_GROUP_NUM:
642 if (copy_to_user(argp, &v->vdpa->ngroups,
643 sizeof(v->vdpa->ngroups)))
644 r = -EFAULT;
645 break;
646 case VHOST_VDPA_GET_AS_NUM:
647 if (copy_to_user(argp, &v->vdpa->nas, sizeof(v->vdpa->nas)))
648 r = -EFAULT;
649 break;
650 case VHOST_SET_LOG_BASE:
651 case VHOST_SET_LOG_FD:
652 r = -ENOIOCTLCMD;
653 break;
654 case VHOST_VDPA_SET_CONFIG_CALL:
655 r = vhost_vdpa_set_config_call(v, argp);
656 break;
657 case VHOST_GET_BACKEND_FEATURES:
658 features = VHOST_VDPA_BACKEND_FEATURES;
659 if (vhost_vdpa_can_suspend(v))
660 features |= BIT_ULL(VHOST_BACKEND_F_SUSPEND);
661 if (copy_to_user(featurep, &features, sizeof(features)))
662 r = -EFAULT;
663 break;
664 case VHOST_VDPA_GET_IOVA_RANGE:
665 r = vhost_vdpa_get_iova_range(v, argp);
666 break;
667 case VHOST_VDPA_GET_CONFIG_SIZE:
668 r = vhost_vdpa_get_config_size(v, argp);
669 break;
670 case VHOST_VDPA_GET_VQS_COUNT:
671 r = vhost_vdpa_get_vqs_count(v, argp);
672 break;
673 case VHOST_VDPA_SUSPEND:
674 r = vhost_vdpa_suspend(v);
675 break;
676 default:
677 r = vhost_dev_ioctl(&v->vdev, cmd, argp);
678 if (r == -ENOIOCTLCMD)
679 r = vhost_vdpa_vring_ioctl(v, cmd, argp);
680 break;
681 }
682
683 mutex_unlock(&d->mutex);
684 return r;
685 }
686
vhost_vdpa_pa_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last)687 static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v,
688 struct vhost_iotlb *iotlb,
689 u64 start, u64 last)
690 {
691 struct vhost_dev *dev = &v->vdev;
692 struct vhost_iotlb_map *map;
693 struct page *page;
694 unsigned long pfn, pinned;
695
696 while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
697 pinned = PFN_DOWN(map->size);
698 for (pfn = PFN_DOWN(map->addr);
699 pinned > 0; pfn++, pinned--) {
700 page = pfn_to_page(pfn);
701 if (map->perm & VHOST_ACCESS_WO)
702 set_page_dirty_lock(page);
703 unpin_user_page(page);
704 }
705 atomic64_sub(PFN_DOWN(map->size), &dev->mm->pinned_vm);
706 vhost_iotlb_map_free(iotlb, map);
707 }
708 }
709
vhost_vdpa_va_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last)710 static void vhost_vdpa_va_unmap(struct vhost_vdpa *v,
711 struct vhost_iotlb *iotlb,
712 u64 start, u64 last)
713 {
714 struct vhost_iotlb_map *map;
715 struct vdpa_map_file *map_file;
716
717 while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
718 map_file = (struct vdpa_map_file *)map->opaque;
719 fput(map_file->file);
720 kfree(map_file);
721 vhost_iotlb_map_free(iotlb, map);
722 }
723 }
724
vhost_vdpa_iotlb_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last)725 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v,
726 struct vhost_iotlb *iotlb,
727 u64 start, u64 last)
728 {
729 struct vdpa_device *vdpa = v->vdpa;
730
731 if (vdpa->use_va)
732 return vhost_vdpa_va_unmap(v, iotlb, start, last);
733
734 return vhost_vdpa_pa_unmap(v, iotlb, start, last);
735 }
736
perm_to_iommu_flags(u32 perm)737 static int perm_to_iommu_flags(u32 perm)
738 {
739 int flags = 0;
740
741 switch (perm) {
742 case VHOST_ACCESS_WO:
743 flags |= IOMMU_WRITE;
744 break;
745 case VHOST_ACCESS_RO:
746 flags |= IOMMU_READ;
747 break;
748 case VHOST_ACCESS_RW:
749 flags |= (IOMMU_WRITE | IOMMU_READ);
750 break;
751 default:
752 WARN(1, "invalidate vhost IOTLB permission\n");
753 break;
754 }
755
756 return flags | IOMMU_CACHE;
757 }
758
vhost_vdpa_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 pa,u32 perm,void * opaque)759 static int vhost_vdpa_map(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
760 u64 iova, u64 size, u64 pa, u32 perm, void *opaque)
761 {
762 struct vhost_dev *dev = &v->vdev;
763 struct vdpa_device *vdpa = v->vdpa;
764 const struct vdpa_config_ops *ops = vdpa->config;
765 u32 asid = iotlb_to_asid(iotlb);
766 int r = 0;
767
768 r = vhost_iotlb_add_range_ctx(iotlb, iova, iova + size - 1,
769 pa, perm, opaque);
770 if (r)
771 return r;
772
773 if (ops->dma_map) {
774 r = ops->dma_map(vdpa, asid, iova, size, pa, perm, opaque);
775 } else if (ops->set_map) {
776 if (!v->in_batch)
777 r = ops->set_map(vdpa, asid, iotlb);
778 } else {
779 r = iommu_map(v->domain, iova, pa, size,
780 perm_to_iommu_flags(perm));
781 }
782 if (r) {
783 vhost_iotlb_del_range(iotlb, iova, iova + size - 1);
784 return r;
785 }
786
787 if (!vdpa->use_va)
788 atomic64_add(PFN_DOWN(size), &dev->mm->pinned_vm);
789
790 return 0;
791 }
792
vhost_vdpa_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size)793 static void vhost_vdpa_unmap(struct vhost_vdpa *v,
794 struct vhost_iotlb *iotlb,
795 u64 iova, u64 size)
796 {
797 struct vdpa_device *vdpa = v->vdpa;
798 const struct vdpa_config_ops *ops = vdpa->config;
799 u32 asid = iotlb_to_asid(iotlb);
800
801 vhost_vdpa_iotlb_unmap(v, iotlb, iova, iova + size - 1);
802
803 if (ops->dma_map) {
804 ops->dma_unmap(vdpa, asid, iova, size);
805 } else if (ops->set_map) {
806 if (!v->in_batch)
807 ops->set_map(vdpa, asid, iotlb);
808 } else {
809 iommu_unmap(v->domain, iova, size);
810 }
811
812 /* If we are in the middle of batch processing, delay the free
813 * of AS until BATCH_END.
814 */
815 if (!v->in_batch && !iotlb->nmaps)
816 vhost_vdpa_remove_as(v, asid);
817 }
818
vhost_vdpa_va_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 uaddr,u32 perm)819 static int vhost_vdpa_va_map(struct vhost_vdpa *v,
820 struct vhost_iotlb *iotlb,
821 u64 iova, u64 size, u64 uaddr, u32 perm)
822 {
823 struct vhost_dev *dev = &v->vdev;
824 u64 offset, map_size, map_iova = iova;
825 struct vdpa_map_file *map_file;
826 struct vm_area_struct *vma;
827 int ret = 0;
828
829 mmap_read_lock(dev->mm);
830
831 while (size) {
832 vma = find_vma(dev->mm, uaddr);
833 if (!vma) {
834 ret = -EINVAL;
835 break;
836 }
837 map_size = min(size, vma->vm_end - uaddr);
838 if (!(vma->vm_file && (vma->vm_flags & VM_SHARED) &&
839 !(vma->vm_flags & (VM_IO | VM_PFNMAP))))
840 goto next;
841
842 map_file = kzalloc(sizeof(*map_file), GFP_KERNEL);
843 if (!map_file) {
844 ret = -ENOMEM;
845 break;
846 }
847 offset = (vma->vm_pgoff << PAGE_SHIFT) + uaddr - vma->vm_start;
848 map_file->offset = offset;
849 map_file->file = get_file(vma->vm_file);
850 ret = vhost_vdpa_map(v, iotlb, map_iova, map_size, uaddr,
851 perm, map_file);
852 if (ret) {
853 fput(map_file->file);
854 kfree(map_file);
855 break;
856 }
857 next:
858 size -= map_size;
859 uaddr += map_size;
860 map_iova += map_size;
861 }
862 if (ret)
863 vhost_vdpa_unmap(v, iotlb, iova, map_iova - iova);
864
865 mmap_read_unlock(dev->mm);
866
867 return ret;
868 }
869
vhost_vdpa_pa_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 uaddr,u32 perm)870 static int vhost_vdpa_pa_map(struct vhost_vdpa *v,
871 struct vhost_iotlb *iotlb,
872 u64 iova, u64 size, u64 uaddr, u32 perm)
873 {
874 struct vhost_dev *dev = &v->vdev;
875 struct page **page_list;
876 unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
877 unsigned int gup_flags = FOLL_LONGTERM;
878 unsigned long npages, cur_base, map_pfn, last_pfn = 0;
879 unsigned long lock_limit, sz2pin, nchunks, i;
880 u64 start = iova;
881 long pinned;
882 int ret = 0;
883
884 /* Limit the use of memory for bookkeeping */
885 page_list = (struct page **) __get_free_page(GFP_KERNEL);
886 if (!page_list)
887 return -ENOMEM;
888
889 if (perm & VHOST_ACCESS_WO)
890 gup_flags |= FOLL_WRITE;
891
892 npages = PFN_UP(size + (iova & ~PAGE_MASK));
893 if (!npages) {
894 ret = -EINVAL;
895 goto free;
896 }
897
898 mmap_read_lock(dev->mm);
899
900 lock_limit = PFN_DOWN(rlimit(RLIMIT_MEMLOCK));
901 if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
902 ret = -ENOMEM;
903 goto unlock;
904 }
905
906 cur_base = uaddr & PAGE_MASK;
907 iova &= PAGE_MASK;
908 nchunks = 0;
909
910 while (npages) {
911 sz2pin = min_t(unsigned long, npages, list_size);
912 pinned = pin_user_pages(cur_base, sz2pin,
913 gup_flags, page_list, NULL);
914 if (sz2pin != pinned) {
915 if (pinned < 0) {
916 ret = pinned;
917 } else {
918 unpin_user_pages(page_list, pinned);
919 ret = -ENOMEM;
920 }
921 goto out;
922 }
923 nchunks++;
924
925 if (!last_pfn)
926 map_pfn = page_to_pfn(page_list[0]);
927
928 for (i = 0; i < pinned; i++) {
929 unsigned long this_pfn = page_to_pfn(page_list[i]);
930 u64 csize;
931
932 if (last_pfn && (this_pfn != last_pfn + 1)) {
933 /* Pin a contiguous chunk of memory */
934 csize = PFN_PHYS(last_pfn - map_pfn + 1);
935 ret = vhost_vdpa_map(v, iotlb, iova, csize,
936 PFN_PHYS(map_pfn),
937 perm, NULL);
938 if (ret) {
939 /*
940 * Unpin the pages that are left unmapped
941 * from this point on in the current
942 * page_list. The remaining outstanding
943 * ones which may stride across several
944 * chunks will be covered in the common
945 * error path subsequently.
946 */
947 unpin_user_pages(&page_list[i],
948 pinned - i);
949 goto out;
950 }
951
952 map_pfn = this_pfn;
953 iova += csize;
954 nchunks = 0;
955 }
956
957 last_pfn = this_pfn;
958 }
959
960 cur_base += PFN_PHYS(pinned);
961 npages -= pinned;
962 }
963
964 /* Pin the rest chunk */
965 ret = vhost_vdpa_map(v, iotlb, iova, PFN_PHYS(last_pfn - map_pfn + 1),
966 PFN_PHYS(map_pfn), perm, NULL);
967 out:
968 if (ret) {
969 if (nchunks) {
970 unsigned long pfn;
971
972 /*
973 * Unpin the outstanding pages which are yet to be
974 * mapped but haven't due to vdpa_map() or
975 * pin_user_pages() failure.
976 *
977 * Mapped pages are accounted in vdpa_map(), hence
978 * the corresponding unpinning will be handled by
979 * vdpa_unmap().
980 */
981 WARN_ON(!last_pfn);
982 for (pfn = map_pfn; pfn <= last_pfn; pfn++)
983 unpin_user_page(pfn_to_page(pfn));
984 }
985 vhost_vdpa_unmap(v, iotlb, start, size);
986 }
987 unlock:
988 mmap_read_unlock(dev->mm);
989 free:
990 free_page((unsigned long)page_list);
991 return ret;
992
993 }
994
vhost_vdpa_process_iotlb_update(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,struct vhost_iotlb_msg * msg)995 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
996 struct vhost_iotlb *iotlb,
997 struct vhost_iotlb_msg *msg)
998 {
999 struct vdpa_device *vdpa = v->vdpa;
1000
1001 if (msg->iova < v->range.first || !msg->size ||
1002 msg->iova > U64_MAX - msg->size + 1 ||
1003 msg->iova + msg->size - 1 > v->range.last)
1004 return -EINVAL;
1005
1006 if (vhost_iotlb_itree_first(iotlb, msg->iova,
1007 msg->iova + msg->size - 1))
1008 return -EEXIST;
1009
1010 if (vdpa->use_va)
1011 return vhost_vdpa_va_map(v, iotlb, msg->iova, msg->size,
1012 msg->uaddr, msg->perm);
1013
1014 return vhost_vdpa_pa_map(v, iotlb, msg->iova, msg->size, msg->uaddr,
1015 msg->perm);
1016 }
1017
vhost_vdpa_process_iotlb_msg(struct vhost_dev * dev,u32 asid,struct vhost_iotlb_msg * msg)1018 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev, u32 asid,
1019 struct vhost_iotlb_msg *msg)
1020 {
1021 struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
1022 struct vdpa_device *vdpa = v->vdpa;
1023 const struct vdpa_config_ops *ops = vdpa->config;
1024 struct vhost_iotlb *iotlb = NULL;
1025 struct vhost_vdpa_as *as = NULL;
1026 int r = 0;
1027
1028 mutex_lock(&dev->mutex);
1029
1030 r = vhost_dev_check_owner(dev);
1031 if (r)
1032 goto unlock;
1033
1034 if (msg->type == VHOST_IOTLB_UPDATE ||
1035 msg->type == VHOST_IOTLB_BATCH_BEGIN) {
1036 as = vhost_vdpa_find_alloc_as(v, asid);
1037 if (!as) {
1038 dev_err(&v->dev, "can't find and alloc asid %d\n",
1039 asid);
1040 r = -EINVAL;
1041 goto unlock;
1042 }
1043 iotlb = &as->iotlb;
1044 } else
1045 iotlb = asid_to_iotlb(v, asid);
1046
1047 if ((v->in_batch && v->batch_asid != asid) || !iotlb) {
1048 if (v->in_batch && v->batch_asid != asid) {
1049 dev_info(&v->dev, "batch id %d asid %d\n",
1050 v->batch_asid, asid);
1051 }
1052 if (!iotlb)
1053 dev_err(&v->dev, "no iotlb for asid %d\n", asid);
1054 r = -EINVAL;
1055 goto unlock;
1056 }
1057
1058 switch (msg->type) {
1059 case VHOST_IOTLB_UPDATE:
1060 r = vhost_vdpa_process_iotlb_update(v, iotlb, msg);
1061 break;
1062 case VHOST_IOTLB_INVALIDATE:
1063 vhost_vdpa_unmap(v, iotlb, msg->iova, msg->size);
1064 break;
1065 case VHOST_IOTLB_BATCH_BEGIN:
1066 v->batch_asid = asid;
1067 v->in_batch = true;
1068 break;
1069 case VHOST_IOTLB_BATCH_END:
1070 if (v->in_batch && ops->set_map)
1071 ops->set_map(vdpa, asid, iotlb);
1072 v->in_batch = false;
1073 if (!iotlb->nmaps)
1074 vhost_vdpa_remove_as(v, asid);
1075 break;
1076 default:
1077 r = -EINVAL;
1078 break;
1079 }
1080 unlock:
1081 mutex_unlock(&dev->mutex);
1082
1083 return r;
1084 }
1085
vhost_vdpa_chr_write_iter(struct kiocb * iocb,struct iov_iter * from)1086 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb,
1087 struct iov_iter *from)
1088 {
1089 struct file *file = iocb->ki_filp;
1090 struct vhost_vdpa *v = file->private_data;
1091 struct vhost_dev *dev = &v->vdev;
1092
1093 return vhost_chr_write_iter(dev, from);
1094 }
1095
vhost_vdpa_alloc_domain(struct vhost_vdpa * v)1096 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v)
1097 {
1098 struct vdpa_device *vdpa = v->vdpa;
1099 const struct vdpa_config_ops *ops = vdpa->config;
1100 struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1101 struct bus_type *bus;
1102 int ret;
1103
1104 /* Device want to do DMA by itself */
1105 if (ops->set_map || ops->dma_map)
1106 return 0;
1107
1108 bus = dma_dev->bus;
1109 if (!bus)
1110 return -EFAULT;
1111
1112 if (!device_iommu_capable(dma_dev, IOMMU_CAP_CACHE_COHERENCY))
1113 return -ENOTSUPP;
1114
1115 v->domain = iommu_domain_alloc(bus);
1116 if (!v->domain)
1117 return -EIO;
1118
1119 ret = iommu_attach_device(v->domain, dma_dev);
1120 if (ret)
1121 goto err_attach;
1122
1123 return 0;
1124
1125 err_attach:
1126 iommu_domain_free(v->domain);
1127 return ret;
1128 }
1129
vhost_vdpa_free_domain(struct vhost_vdpa * v)1130 static void vhost_vdpa_free_domain(struct vhost_vdpa *v)
1131 {
1132 struct vdpa_device *vdpa = v->vdpa;
1133 struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1134
1135 if (v->domain) {
1136 iommu_detach_device(v->domain, dma_dev);
1137 iommu_domain_free(v->domain);
1138 }
1139
1140 v->domain = NULL;
1141 }
1142
vhost_vdpa_set_iova_range(struct vhost_vdpa * v)1143 static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
1144 {
1145 struct vdpa_iova_range *range = &v->range;
1146 struct vdpa_device *vdpa = v->vdpa;
1147 const struct vdpa_config_ops *ops = vdpa->config;
1148
1149 if (ops->get_iova_range) {
1150 *range = ops->get_iova_range(vdpa);
1151 } else if (v->domain && v->domain->geometry.force_aperture) {
1152 range->first = v->domain->geometry.aperture_start;
1153 range->last = v->domain->geometry.aperture_end;
1154 } else {
1155 range->first = 0;
1156 range->last = ULLONG_MAX;
1157 }
1158 }
1159
vhost_vdpa_cleanup(struct vhost_vdpa * v)1160 static void vhost_vdpa_cleanup(struct vhost_vdpa *v)
1161 {
1162 struct vhost_vdpa_as *as;
1163 u32 asid;
1164
1165 vhost_dev_cleanup(&v->vdev);
1166 kfree(v->vdev.vqs);
1167
1168 for (asid = 0; asid < v->vdpa->nas; asid++) {
1169 as = asid_to_as(v, asid);
1170 if (as)
1171 vhost_vdpa_remove_as(v, asid);
1172 }
1173 }
1174
vhost_vdpa_open(struct inode * inode,struct file * filep)1175 static int vhost_vdpa_open(struct inode *inode, struct file *filep)
1176 {
1177 struct vhost_vdpa *v;
1178 struct vhost_dev *dev;
1179 struct vhost_virtqueue **vqs;
1180 int r, opened;
1181 u32 i, nvqs;
1182
1183 v = container_of(inode->i_cdev, struct vhost_vdpa, cdev);
1184
1185 opened = atomic_cmpxchg(&v->opened, 0, 1);
1186 if (opened)
1187 return -EBUSY;
1188
1189 nvqs = v->nvqs;
1190 r = vhost_vdpa_reset(v);
1191 if (r)
1192 goto err;
1193
1194 vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
1195 if (!vqs) {
1196 r = -ENOMEM;
1197 goto err;
1198 }
1199
1200 dev = &v->vdev;
1201 for (i = 0; i < nvqs; i++) {
1202 vqs[i] = &v->vqs[i];
1203 vqs[i]->handle_kick = handle_vq_kick;
1204 }
1205 vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
1206 vhost_vdpa_process_iotlb_msg);
1207
1208 r = vhost_vdpa_alloc_domain(v);
1209 if (r)
1210 goto err_alloc_domain;
1211
1212 vhost_vdpa_set_iova_range(v);
1213
1214 filep->private_data = v;
1215
1216 return 0;
1217
1218 err_alloc_domain:
1219 vhost_vdpa_cleanup(v);
1220 err:
1221 atomic_dec(&v->opened);
1222 return r;
1223 }
1224
vhost_vdpa_clean_irq(struct vhost_vdpa * v)1225 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v)
1226 {
1227 u32 i;
1228
1229 for (i = 0; i < v->nvqs; i++)
1230 vhost_vdpa_unsetup_vq_irq(v, i);
1231 }
1232
vhost_vdpa_release(struct inode * inode,struct file * filep)1233 static int vhost_vdpa_release(struct inode *inode, struct file *filep)
1234 {
1235 struct vhost_vdpa *v = filep->private_data;
1236 struct vhost_dev *d = &v->vdev;
1237
1238 mutex_lock(&d->mutex);
1239 filep->private_data = NULL;
1240 vhost_vdpa_clean_irq(v);
1241 vhost_vdpa_reset(v);
1242 vhost_dev_stop(&v->vdev);
1243 vhost_vdpa_free_domain(v);
1244 vhost_vdpa_config_put(v);
1245 vhost_vdpa_cleanup(v);
1246 mutex_unlock(&d->mutex);
1247
1248 atomic_dec(&v->opened);
1249 complete(&v->completion);
1250
1251 return 0;
1252 }
1253
1254 #ifdef CONFIG_MMU
vhost_vdpa_fault(struct vm_fault * vmf)1255 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf)
1256 {
1257 struct vhost_vdpa *v = vmf->vma->vm_file->private_data;
1258 struct vdpa_device *vdpa = v->vdpa;
1259 const struct vdpa_config_ops *ops = vdpa->config;
1260 struct vdpa_notification_area notify;
1261 struct vm_area_struct *vma = vmf->vma;
1262 u16 index = vma->vm_pgoff;
1263
1264 notify = ops->get_vq_notification(vdpa, index);
1265
1266 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1267 if (remap_pfn_range(vma, vmf->address & PAGE_MASK,
1268 PFN_DOWN(notify.addr), PAGE_SIZE,
1269 vma->vm_page_prot))
1270 return VM_FAULT_SIGBUS;
1271
1272 return VM_FAULT_NOPAGE;
1273 }
1274
1275 static const struct vm_operations_struct vhost_vdpa_vm_ops = {
1276 .fault = vhost_vdpa_fault,
1277 };
1278
vhost_vdpa_mmap(struct file * file,struct vm_area_struct * vma)1279 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma)
1280 {
1281 struct vhost_vdpa *v = vma->vm_file->private_data;
1282 struct vdpa_device *vdpa = v->vdpa;
1283 const struct vdpa_config_ops *ops = vdpa->config;
1284 struct vdpa_notification_area notify;
1285 unsigned long index = vma->vm_pgoff;
1286
1287 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1288 return -EINVAL;
1289 if ((vma->vm_flags & VM_SHARED) == 0)
1290 return -EINVAL;
1291 if (vma->vm_flags & VM_READ)
1292 return -EINVAL;
1293 if (index > 65535)
1294 return -EINVAL;
1295 if (!ops->get_vq_notification)
1296 return -ENOTSUPP;
1297
1298 /* To be safe and easily modelled by userspace, We only
1299 * support the doorbell which sits on the page boundary and
1300 * does not share the page with other registers.
1301 */
1302 notify = ops->get_vq_notification(vdpa, index);
1303 if (notify.addr & (PAGE_SIZE - 1))
1304 return -EINVAL;
1305 if (vma->vm_end - vma->vm_start != notify.size)
1306 return -ENOTSUPP;
1307
1308 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1309 vma->vm_ops = &vhost_vdpa_vm_ops;
1310 return 0;
1311 }
1312 #endif /* CONFIG_MMU */
1313
1314 static const struct file_operations vhost_vdpa_fops = {
1315 .owner = THIS_MODULE,
1316 .open = vhost_vdpa_open,
1317 .release = vhost_vdpa_release,
1318 .write_iter = vhost_vdpa_chr_write_iter,
1319 .unlocked_ioctl = vhost_vdpa_unlocked_ioctl,
1320 #ifdef CONFIG_MMU
1321 .mmap = vhost_vdpa_mmap,
1322 #endif /* CONFIG_MMU */
1323 .compat_ioctl = compat_ptr_ioctl,
1324 };
1325
vhost_vdpa_release_dev(struct device * device)1326 static void vhost_vdpa_release_dev(struct device *device)
1327 {
1328 struct vhost_vdpa *v =
1329 container_of(device, struct vhost_vdpa, dev);
1330
1331 ida_simple_remove(&vhost_vdpa_ida, v->minor);
1332 kfree(v->vqs);
1333 kfree(v);
1334 }
1335
vhost_vdpa_probe(struct vdpa_device * vdpa)1336 static int vhost_vdpa_probe(struct vdpa_device *vdpa)
1337 {
1338 const struct vdpa_config_ops *ops = vdpa->config;
1339 struct vhost_vdpa *v;
1340 int minor;
1341 int i, r;
1342
1343 /* We can't support platform IOMMU device with more than 1
1344 * group or as
1345 */
1346 if (!ops->set_map && !ops->dma_map &&
1347 (vdpa->ngroups > 1 || vdpa->nas > 1))
1348 return -EOPNOTSUPP;
1349
1350 v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1351 if (!v)
1352 return -ENOMEM;
1353
1354 minor = ida_simple_get(&vhost_vdpa_ida, 0,
1355 VHOST_VDPA_DEV_MAX, GFP_KERNEL);
1356 if (minor < 0) {
1357 kfree(v);
1358 return minor;
1359 }
1360
1361 atomic_set(&v->opened, 0);
1362 v->minor = minor;
1363 v->vdpa = vdpa;
1364 v->nvqs = vdpa->nvqs;
1365 v->virtio_id = ops->get_device_id(vdpa);
1366
1367 device_initialize(&v->dev);
1368 v->dev.release = vhost_vdpa_release_dev;
1369 v->dev.parent = &vdpa->dev;
1370 v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor);
1371 v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue),
1372 GFP_KERNEL);
1373 if (!v->vqs) {
1374 r = -ENOMEM;
1375 goto err;
1376 }
1377
1378 r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor);
1379 if (r)
1380 goto err;
1381
1382 cdev_init(&v->cdev, &vhost_vdpa_fops);
1383 v->cdev.owner = THIS_MODULE;
1384
1385 r = cdev_device_add(&v->cdev, &v->dev);
1386 if (r)
1387 goto err;
1388
1389 init_completion(&v->completion);
1390 vdpa_set_drvdata(vdpa, v);
1391
1392 for (i = 0; i < VHOST_VDPA_IOTLB_BUCKETS; i++)
1393 INIT_HLIST_HEAD(&v->as[i]);
1394
1395 return 0;
1396
1397 err:
1398 put_device(&v->dev);
1399 ida_simple_remove(&vhost_vdpa_ida, v->minor);
1400 return r;
1401 }
1402
vhost_vdpa_remove(struct vdpa_device * vdpa)1403 static void vhost_vdpa_remove(struct vdpa_device *vdpa)
1404 {
1405 struct vhost_vdpa *v = vdpa_get_drvdata(vdpa);
1406 int opened;
1407
1408 cdev_device_del(&v->cdev, &v->dev);
1409
1410 do {
1411 opened = atomic_cmpxchg(&v->opened, 0, 1);
1412 if (!opened)
1413 break;
1414 wait_for_completion(&v->completion);
1415 } while (1);
1416
1417 put_device(&v->dev);
1418 }
1419
1420 static struct vdpa_driver vhost_vdpa_driver = {
1421 .driver = {
1422 .name = "vhost_vdpa",
1423 },
1424 .probe = vhost_vdpa_probe,
1425 .remove = vhost_vdpa_remove,
1426 };
1427
vhost_vdpa_init(void)1428 static int __init vhost_vdpa_init(void)
1429 {
1430 int r;
1431
1432 r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX,
1433 "vhost-vdpa");
1434 if (r)
1435 goto err_alloc_chrdev;
1436
1437 r = vdpa_register_driver(&vhost_vdpa_driver);
1438 if (r)
1439 goto err_vdpa_register_driver;
1440
1441 return 0;
1442
1443 err_vdpa_register_driver:
1444 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1445 err_alloc_chrdev:
1446 return r;
1447 }
1448 module_init(vhost_vdpa_init);
1449
vhost_vdpa_exit(void)1450 static void __exit vhost_vdpa_exit(void)
1451 {
1452 vdpa_unregister_driver(&vhost_vdpa_driver);
1453 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1454 }
1455 module_exit(vhost_vdpa_exit);
1456
1457 MODULE_VERSION("0.0.1");
1458 MODULE_LICENSE("GPL v2");
1459 MODULE_AUTHOR("Intel Corporation");
1460 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio");
1461